Is there a Hopfield pattern recognition with similarity coefficient determination?

Question

Machine learning

In general, the essence of the question is whether it is possible to obtain a coefficient (from .0 to 1.0 or whatever) when recognizing an image, which, in the opinion of the program, would determine the similarity (similarity) of the image presented to the program with any image from the training sample? (with the possibility of determining the similarity coefficients with episodes from the training sample?)
Is there a classic Hopfield network in python, is it possible to set up the algorithm in such a way that when determining the image it would be possible to obtain a similarity coefficient?

Answer 1

Two options come to mind:
1) Head-on: train one network (something simple from convolutional networks, like VGG) on the entire dataset, take features from the penultimate layer for each of the images that need to be compared, calculate the cosine distance between the received vectors . It is written in python in 50-100 lines.
2) If you want to get confused: siamese networks .